Case Study: Voicebox achieves improved ASR accuracy and faster time-to-market with Databricks

Voicebox - Customer Case Study

Voicebox is a leader in conversational AI and secure speech recognition for IoT voice assistants. They struggled with accuracy because rapidly changing data (points of interest, addresses, music, etc.) needed constant normalization and aliasing for ASR/NLU models, and legacy manual pipelines caused high latency and slow time-to-market.

Using Databricks, Voicebox automated production pipelines—including criteria pipelines to keep models up to date and a latency pipeline to accelerate data processing—feeding reliable data into their deep learning models. The result was reduced engineering complexity, faster iteration, expanded market opportunities beyond automotive, and increased revenue.

Voicebox

Peyvand Khademi

Director, Data Platform and Services